Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A method for responding to a natural language communication, the method comprising: receiving, by a server with a uniform resource identifier (URI), a hypertext transfer protocol (HTTP) post call message, wherein the HTTP post call message is directed to the URI from a messaging application server, wherein the messaging application server receives a first user message comprising a message from a user via a messaging application on a device of the user, and wherein the HTTP post call message includes content from the first user message; calculating a first vector for the content, wherein the first vector is a distributed, mathematical representation of the content from the first user message in a vector space defined by a language model such that semantically similar texts are mapped to nearby points, wherein the first vector includes a plurality of first elements, and wherein each first element of the plurality of first elements characterizes the content from the first user message according to the language model; identifying a plurality of second vectors from a plurality of predefined intent vectors in a knowledge pack, the knowledge pack comprising the plurality of predefined intent vectors in the vector space defined by the language model, each predefined intent vector having a known intent, wherein each second vector is a distributed, mathematical representation of one or more example messages, respectively, wherein each second vector includes a subset of a plurality of second elements, and wherein each subset of second elements characterizes the one or more example messages of the associated second vector according to the language model; comparing the first vector to each of the plurality of second vectors; determining an intent of the content from the first user message based on the comparison; generating a response to the first user message based on the determined intent; and sending the response to the messaging application server.
2. The method of claim 1 , wherein comparing the first vector to each of the plurality of second vectors comprises: calculating a Euclidian distance between the first vector and each of the plurality of second vectors to generate a plurality of distances, and wherein determining an intent of the content from the first user message comprises: identifying a distance of the plurality of distances that is below a threshold; and determining the intent of the content from the known intent associated with the second vector associated with the distance.
This invention relates to natural language processing and intent recognition in user messages. The problem addressed is accurately identifying the intent of user messages in applications like chatbots or virtual assistants, where messages may be ambiguous or context-dependent. The method involves processing a first user message to generate a first vector representing its semantic content. This vector is compared to a plurality of second vectors, each associated with a known intent. The comparison is performed by calculating the Euclidean distance between the first vector and each second vector, producing a set of distance values. The intent of the user message is determined by identifying the smallest distance below a predefined threshold and associating the user message with the known intent linked to the corresponding second vector. The method ensures that only semantically similar vectors with distances below the threshold are considered, improving intent recognition accuracy. The approach leverages vector embeddings and distance metrics to handle variations in user input while maintaining reliable intent classification. This technique is particularly useful in systems requiring precise interpretation of user queries or commands.
3. The method of claim 1 , wherein comparing the first vector to each of the plurality of second vectors comprises: calculating a cosine similarity between the first vector and each of the plurality of second vectors to generate a plurality of similarity values, wherein calculating the cosine similarity comprises calculating the cosine of an angle between the first vector and each of the plurality of second vectors, and wherein determining an intent of the content from the first user message comprises: identifying a similarity value of the plurality of similarity values that is below a threshold distance from a desired similarity value; and determining the intent of the content from the known intent associated with the second vector associated with the similarity value.
This invention relates to natural language processing and intent recognition in user messages. The problem addressed is accurately determining the intent of a user's message by comparing it to a set of predefined intents represented as vectors. The method involves generating a vector representation of the user's message and comparing it to multiple predefined intent vectors using cosine similarity. Cosine similarity measures the angle between vectors, with smaller angles indicating higher similarity. The method calculates the cosine of the angle between the user's message vector and each predefined intent vector, producing a set of similarity values. To determine the intent, the method identifies the similarity value closest to a desired threshold and associates the user's message with the known intent linked to the corresponding predefined vector. This approach improves intent recognition by leveraging vector-based similarity metrics to match user input to predefined categories.
4. The method of claim 1 , wherein a first element of the plurality of first elements is not mutually exclusive with a second element of the plurality of first elements.
This invention relates to systems and methods for managing overlapping or non-mutually exclusive elements within a set of data elements. The problem addressed is the need to efficiently handle scenarios where elements in a dataset are not mutually exclusive, meaning that a single element can belong to multiple categories or groups simultaneously. This is particularly relevant in applications such as data classification, tagging, or hierarchical categorization, where traditional mutually exclusive models fail to capture the complexity of real-world relationships. The invention provides a method for processing a plurality of first elements, where at least one first element is not mutually exclusive with another first element. This means that an element can be associated with multiple categories or attributes without conflict. The method ensures that overlapping or shared elements are properly managed, allowing for more flexible and accurate data representation. The approach may involve techniques such as multi-label classification, hierarchical tagging, or probabilistic modeling to handle non-mutually exclusive relationships. The system may also include mechanisms to resolve conflicts or ambiguities when elements overlap, ensuring consistency in data processing. This method is particularly useful in fields like machine learning, database management, and information retrieval, where accurate and flexible categorization is essential. The invention improves upon prior art by providing a more adaptable framework for handling non-mutually exclusive elements, enhancing data accuracy and usability.
5. The method of claim 1 , further comprising: identifying an entity in the content from the first user message, wherein the entity is predefined to be associated with a first known intent from the known intents of the second vectors, and wherein determining the intent of the content is further based on the entity.
This invention relates to natural language processing (NLP) systems for intent recognition in user messages, particularly in conversational interfaces like chatbots or virtual assistants. The problem addressed is accurately determining user intent from ambiguous or incomplete input, which is critical for effective automated responses. The method involves processing a first user message to generate a first vector representation of the message content. This vector is compared against a set of second vectors, each representing known intents, to identify the most likely intent. The comparison is enhanced by identifying predefined entities within the message content. These entities are linked to specific known intents, providing additional context to refine intent recognition. For example, if a user mentions "flight" (a predefined entity), the system may prioritize travel-related intents over others. The final intent determination combines both the vector similarity and entity-based context, improving accuracy in ambiguous scenarios. This approach helps systems better understand user requests, even when input is incomplete or lacks clear intent indicators.
6. The method of claim 5 , further comprising: receiving the knowledge pack, wherein a first known intent of the known intents of the second vectors includes one or more entities, wherein the knowledge pack is associated with a domain, and wherein the domain includes the known intents.
This invention relates to natural language processing (NLP) systems that use knowledge packs to improve intent recognition. The problem addressed is the difficulty in accurately identifying user intents in conversational systems, particularly when those intents involve specific entities or domain-specific knowledge. The solution involves enhancing intent recognition by incorporating structured knowledge packs that contain predefined intents and associated entities within a specific domain. The method involves receiving a knowledge pack that is linked to a particular domain. This knowledge pack includes known intents, where at least one of these intents is associated with one or more entities. The domain itself encompasses these known intents, allowing the system to leverage domain-specific knowledge to improve intent classification. The knowledge pack may be used to train or refine an NLP model, enabling it to better recognize user queries that match the known intents and entities within the domain. This approach ensures that the system can handle domain-specific language patterns and entities more effectively, leading to more accurate and contextually relevant responses. The method may also involve processing input data, such as user queries, to extract features and compare them against the known intents and entities in the knowledge pack to determine the most likely intent.
7. The method of claim 1 , wherein a first second vector of the plurality of second vectors is associated with a first known intent, wherein a second vector of the plurality of second vectors is associated with a second known intent, and wherein the first known intent is different than the second known intent.
This invention relates to natural language processing (NLP) systems that classify user inputs into different intents. The problem addressed is the difficulty in accurately distinguishing between multiple intents in user queries, particularly when the intents are semantically similar. The solution involves using vector representations of known intents to improve intent classification. The system processes user inputs by converting them into vector representations. These vectors are compared against a plurality of pre-defined second vectors, each associated with a distinct known intent. The second vectors are derived from training data where each intent is represented by one or more vectors. The system ensures that at least two of these second vectors correspond to different known intents, allowing the system to differentiate between them. By comparing the user input vector against these second vectors, the system can classify the input into the most likely intent. This approach enhances intent classification accuracy by leveraging multiple vector representations for each intent, reducing ambiguity and improving the system's ability to handle nuanced or overlapping intents. The method is particularly useful in applications like virtual assistants, chatbots, and automated customer service systems where precise intent recognition is critical.
8. A system for responding to a natural language communication, the system comprising: one or more processors; and a non-transitory computer-readable medium containing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations including: receive a hypertext transfer protocol (HTTP) post call message, wherein the system has a uniform resource identifier (URI), wherein the HTTP post call message is directed to the URI from a messaging application server, wherein the messaging application server receives a first user message comprising a message from a user via a messaging application on a device of the user, and wherein the HTTP post call message includes content from the first user message; calculate a first vector for the content, wherein the first vector is a distributed, mathematical representation of the content from the first user message in a vector space defined by a language model such that semantically similar texts are mapped to nearby points, wherein the first vector includes a plurality of first elements, wherein each first element of the plurality of first elements characterizes the content from the first user message according to the language model; identify a plurality of second vectors from a plurality of predefined intent vectors in a knowledge pack, the knowledge pack comprising the plurality of predefined intent vectors in the vector space defined by the language model, each predefined intent vector having a known intent, wherein each second vector is a distributed, mathematical representation of one or more example messages, respectively, wherein each second vector includes a subset of a plurality of second elements, wherein each subset of second elements characterizes the one or more example messages of the associated second vector according to the language model; compare the first vector to each of the plurality of second vectors; determine an intent of the content from the first user message based on the comparison; generating a response to the first user message based on the determined intent; and send the response to the messaging application server.
A system processes natural language communications by analyzing user messages received via a messaging application. The system receives an HTTP POST call from a messaging application server, which includes a user's message content. The system converts this content into a vector representation using a language model, where semantically similar texts are mapped to nearby points in a vector space. Each element of the vector characterizes the message content according to the language model. The system then compares this vector to a set of predefined intent vectors stored in a knowledge pack. Each intent vector represents one or more example messages with known intents, also encoded as vectors in the same vector space. By comparing the user's message vector to these intent vectors, the system determines the most likely intent of the user's message. Based on the identified intent, the system generates an appropriate response and sends it back to the messaging application server for delivery to the user. This approach enables automated, context-aware responses to natural language inputs in messaging applications.
9. The system of claim 8 , wherein comparing the first vector to each of the plurality of second vectors comprises: calculating a Euclidian distance between the first vector and each of the plurality of second vectors to generate a plurality of distances, and wherein determining an intent of the content from the first user message comprises: identifying a distance of the plurality of distances that is below a threshold; and determining the intent of the content from the known intent associated with the second vector associated with the distance.
A system for analyzing user messages to determine intent involves processing natural language input to identify user objectives. The system converts a first user message into a first vector representation using a neural network model, such as a transformer-based model, which encodes the message into a numerical vector. This vector is then compared to a plurality of second vectors, each representing a known intent and stored in a database. The comparison is performed by calculating the Euclidean distance between the first vector and each of the second vectors, generating a set of distance values. The system then identifies the smallest distance below a predefined threshold, indicating the closest match between the user message and a known intent. The intent associated with the second vector corresponding to this smallest distance is selected as the determined intent of the user message. This approach enables automated classification of user queries into predefined categories, improving efficiency in applications like customer support, chatbots, or search engines. The system may also include preprocessing steps to clean or normalize the input message before vectorization.
10. The system of claim 8 , wherein comparing the first vector to each of the plurality of second vectors comprises: calculating a cosine similarity between the first vector and each of the plurality of second vectors to generate a plurality of similarity values, wherein calculating the cosine similarity comprises calculating the cosine of an angle between the first vector and each of the plurality of second vectors, and wherein determining an intent of the content from the first user message comprises: identifying a similarity value of the plurality of similarity values that is below a threshold distance from a desired similarity value; and determining the intent of the content from the known intent associated with the second vector associated with the similarity value.
This invention relates to a system for determining the intent of user messages by comparing vector representations of the messages to a database of known intents. The system addresses the challenge of accurately interpreting user input in natural language processing (NLP) applications, where messages may vary in phrasing but convey similar meanings. The system generates a first vector representing the content of a user message and compares it to a plurality of second vectors, each associated with a known intent. The comparison involves calculating the cosine similarity between the first vector and each second vector, which measures the angle between the vectors in a high-dimensional space. A higher cosine similarity indicates greater semantic similarity. The system then identifies the intent of the user message by selecting the known intent associated with the second vector that has a similarity value closest to a predefined threshold. This approach enables the system to classify user messages into predefined intent categories, improving the accuracy of NLP applications such as chatbots, virtual assistants, and automated customer support systems. The system may also include preprocessing steps to normalize the user message and generate the first vector, as well as post-processing steps to refine the intent determination.
11. The system of claim 8 , wherein a first element of the plurality of first elements is not mutually exclusive with a second element of the plurality of first elements.
A system for managing overlapping or non-mutually exclusive elements within a classification or categorization framework. The system addresses the challenge of organizing data or items where elements may belong to multiple categories simultaneously, rather than being strictly exclusive. This is particularly useful in scenarios where traditional hierarchical or mutually exclusive classification systems fail to capture real-world relationships, such as in knowledge management, taxonomy design, or data tagging systems. The system includes a plurality of first elements, where at least one first element is not mutually exclusive with another first element. This means that an item or data point can be associated with multiple first elements without conflict, allowing for more flexible and accurate categorization. The system may also include a plurality of second elements, where each second element is associated with one or more first elements. These second elements may represent subcategories, tags, or attributes that further refine the classification structure. The system may further include a processor configured to process and manage these relationships, ensuring that the non-mutually exclusive nature of the first elements is preserved while maintaining logical consistency within the classification framework. This approach enables more nuanced and adaptable categorization, improving data retrieval, analysis, and decision-making in complex environments.
12. The system of claim 8 , wherein the non-transitory computer-readable medium further includes instructions that, when executed by the one or more processors, cause the one or more processors to perform operations including: identify an entity in the content from the first user message, wherein the entity is predefined to be associated with a first known intent from the known intents of the second vectors, and wherein determining the intent of the content is further based on the entity.
This system relates to natural language processing and intent recognition in conversational interfaces. The problem addressed is accurately determining user intent in messages, particularly when the message contains entities that are predefined to correspond to specific intents. The system processes user messages by converting them into vector representations and comparing them to known intent vectors to identify the most likely intent. A key feature is the ability to recognize predefined entities within the message content. When such an entity is detected, it is associated with a known intent, which influences the overall intent determination process. This ensures that messages containing specific entities are correctly classified according to their predefined intents, improving the accuracy of intent recognition in conversational systems. The system leverages machine learning models to analyze the message content and entity associations, enhancing the reliability of intent detection in automated interactions.
13. The system of claim 12 , wherein the non-transitory computer-readable medium further includes instructions that, when executed by the one or more processors, cause the one or more processors to perform operations including: receive the knowledge pack, wherein a first known intent of the known intents of the second vectors includes one or more entities, wherein the knowledge pack is associated with a domain, and wherein the domain includes the known intents.
This invention relates to a system for processing and utilizing knowledge packs in a domain-specific context. The system is designed to enhance natural language understanding by leveraging structured knowledge representations, particularly in applications like virtual assistants, chatbots, or automated customer service systems. The core problem addressed is the efficient retrieval and application of domain-specific knowledge to improve intent recognition and entity extraction from user inputs. The system includes a non-transitory computer-readable medium storing instructions that, when executed by one or more processors, enable the system to receive a knowledge pack. This knowledge pack contains second vectors representing known intents, where at least one of these intents includes one or more entities. The knowledge pack is associated with a specific domain, which encompasses the known intents relevant to that domain. The system processes this structured knowledge to improve the accuracy and context-awareness of intent recognition, ensuring that user inputs are correctly interpreted within the domain's context. This approach allows for more precise and relevant responses by the system, particularly in specialized fields where domain-specific terminology and relationships are critical. The system's ability to dynamically incorporate and utilize knowledge packs enhances its adaptability to different domains without requiring extensive retraining.
14. A computer-program product tangibly embodied in a non-transitory machine-readable storage medium, including instructions that, when executed by one or more processors, cause the one or more processors to: receiving, by a server with a uniform resource identifier (URI), a hypertext transfer protocol (HTTP) post call message, wherein the HTTP post call message is directed to the URI from a messaging application server, wherein the messaging application server receives a first user message comprising a message from a user via a messaging application on a device of the user, and wherein the HTTP post call message includes content from the first user message; calculating a first vector for the content, wherein the first vector is a distributed, mathematical representation of the content from the first user message in a vector space defined by a language model such that semantically similar texts are mapped to nearby points, wherein the first vector includes a plurality of first elements, wherein each first element of the plurality of first elements characterizes the content from the first user message according to the language model; identifying a plurality of second vectors from a plurality of predefined intent vectors in a knowledge pack, the knowledge pack comprising the a plurality of predefined intent vectors in the vector space defined by the language model, each predefined intent vector having a known intent, wherein each second vector is a distributed, mathematical representation of one or more example messages, respectively, wherein each second vector includes a subset of a plurality of second elements, wherein each subset of second elements characterizes the one or more example messages of the associated second vector according to the language model; comparing the first vector to each of the plurality of second vectors; determining an intent of the content from the first user message based on the comparison; generating a response to the first user message based on the determined intent; and sending the response to the messaging application server.
This invention relates to natural language processing (NLP) and intent recognition in messaging systems. The problem addressed is accurately interpreting user messages in messaging applications to determine their intent and generate appropriate responses. The solution involves a server system that processes HTTP POST messages containing user-generated content from a messaging application. The server receives a user message, extracts its content, and converts it into a mathematical vector representation using a language model. This vector is a distributed representation in a vector space where semantically similar texts are mapped to nearby points. Each element of the vector characterizes the content according to the language model. The system then compares this vector to a set of predefined intent vectors stored in a knowledge pack. Each predefined intent vector represents one or more example messages with known intents and is also a distributed mathematical representation in the same vector space. By comparing the user message vector to these predefined intent vectors, the system determines the most likely intent of the user message. Based on this intent, the system generates an appropriate response and sends it back to the messaging application server for delivery to the user. This approach enables automated, context-aware responses in messaging applications by leveraging semantic similarity in vector space representations.
15. The computer-program product of claim 14 , wherein comparing the first vector to each of the plurality of second vectors comprises: calculating a Euclidian distance between the first vector and each of the plurality of second vectors to generate a plurality of distances, and wherein determining an intent of the content from the first user message comprises: identifying a distance of the plurality of distances that is below a threshold; and determining the intent of the content from the known intent associated with the second vector associated with the distance.
This invention relates to natural language processing (NLP) and intent recognition in user messages. The problem addressed is accurately determining the intent of user messages in applications like chatbots or virtual assistants, where messages may be ambiguous or context-dependent. The solution involves comparing a vector representation of a user message to a set of pre-labeled vectors representing known intents, using Euclidean distance as a similarity metric. The system generates a vector representation of a user message and compares it to a plurality of pre-existing vectors, each associated with a known intent. The comparison is performed by calculating the Euclidean distance between the user message vector and each of the pre-labeled vectors. The intent of the user message is then determined by identifying the smallest Euclidean distance below a predefined threshold and associating the user message with the intent of the closest pre-labeled vector. This approach leverages vector embeddings and distance metrics to classify user input into predefined intent categories, improving accuracy in intent recognition tasks. The method is particularly useful in automated customer service, voice assistants, and other interactive systems where understanding user intent is critical.
16. The computer-program product of claim 14 , wherein comparing the first vector to each of the plurality of second vectors comprises: calculating a cosine similarity between the first vector and each of the plurality of second vectors to generate a plurality of similarity values, wherein calculating the cosine similarity comprises calculating the cosine of an angle between the first vector and each of the plurality of second vectors, and wherein determining an intent of the content from the first user message comprises: identifying a similarity value of the plurality of similarity values that is below a threshold distance from a desired similarity value; and determining the intent of the content from the known intent associated with the second vector associated with the similarity value.
This invention relates to natural language processing and intent recognition in user messages. The problem addressed is accurately determining the intent of user messages by comparing them to a set of predefined intents represented as vectors. The solution involves calculating cosine similarity between a vector representing the user message and a plurality of vectors representing known intents. The cosine similarity measures the angle between vectors, with smaller angles indicating higher similarity. The system identifies a similarity value that is within a threshold distance of a desired similarity value and then determines the intent of the user message based on the known intent associated with the vector that produced that similarity value. This approach improves intent recognition by leveraging vector-based comparisons and similarity thresholds to enhance accuracy. The method is implemented as a computer program product, ensuring efficient and scalable processing of user messages for intent classification.
17. The computer-program product of claim 14 , wherein a first element of the plurality of first elements is not mutually exclusive with a second element of the plurality of first elements.
This invention relates to a computer-program product for managing and processing data elements, particularly in systems where elements may overlap or share attributes. The problem addressed is the need to handle non-mutually exclusive elements efficiently, where multiple elements can coexist or share characteristics without conflict. The invention provides a method to process a plurality of first elements, where at least one first element is not mutually exclusive with another first element, meaning they can overlap or share attributes without restriction. The system includes a processor and a memory storing instructions that, when executed, cause the processor to perform operations such as receiving, processing, and storing these non-mutually exclusive elements. The invention also involves generating outputs based on these elements, ensuring that overlapping or shared attributes are correctly managed. The solution is particularly useful in applications like data classification, where elements may belong to multiple categories simultaneously, or in systems requiring flexible attribute assignment. The invention ensures that non-mutually exclusive elements are processed without errors, maintaining data integrity and accuracy in overlapping scenarios.
18. The computer-program product of claim 14 , wherein the instructions, when executed by the one or more processors, further cause the one or more processors to: identify an entity in the content from the first user message, wherein the entity is predefined to be associated with a first known intent from the known intents of the second vectors, and wherein determining the intent of the content is further based on the entity.
This invention relates to natural language processing (NLP) systems for intent recognition in user messages. The problem addressed is accurately determining the intent behind user inputs, particularly when the content contains entities that may influence the interpretation. The system processes user messages by converting them into vector representations, which are then compared to predefined intent vectors to identify the most likely intent. The system further enhances intent recognition by detecting predefined entities within the message content. These entities are linked to specific known intents, allowing the system to refine its intent determination by considering both the vector-based analysis and the contextual clues provided by the entities. This dual approach improves accuracy in intent classification, especially in scenarios where ambiguous or complex language is used. The system is designed to operate within a computing environment, leveraging machine learning or rule-based methods to map entities to their associated intents. The overall goal is to enable more precise and context-aware interactions in applications such as chatbots, virtual assistants, or customer support systems.
19. The computer-program product of claim 18 , wherein the instructions, when executed by the one or more processors, further cause the one or more processors to: receive the knowledge pack, wherein a first known intent of the known intents of the second vectors includes one or more entities, wherein the knowledge pack is associated with a domain, and wherein the domain includes the known intents.
This invention relates to natural language processing (NLP) systems that use knowledge packs to improve intent recognition in user queries. The problem addressed is the difficulty in accurately identifying user intents in conversational systems, particularly when queries contain entities or domain-specific terms. The solution involves a computer-program product that processes user input by comparing it to pre-existing knowledge packs. These knowledge packs contain known intents, each associated with one or more entities and organized within a specific domain. The system receives a knowledge pack that includes these known intents, where at least one intent is linked to one or more entities. The domain associated with the knowledge pack defines the scope of these intents, ensuring that the system can accurately map user queries to the correct domain-specific intents. This approach enhances the system's ability to understand and respond to user queries by leveraging structured knowledge packs, improving intent recognition accuracy in domain-specific applications. The invention is particularly useful in applications like virtual assistants, customer service bots, and other conversational AI systems where precise intent detection is critical.
20. The computer-program product of claim 14 , wherein a first second vector of the plurality of second vectors is associated with a first known intent, wherein a second vector of the plurality of second vectors is associated with a second known intent, and wherein the first known intent is different than the second known intent.
This invention relates to natural language processing (NLP) systems that classify user inputs into different intents. The problem addressed is the challenge of accurately distinguishing between multiple intents in user queries, particularly when the inputs are ambiguous or contextually similar. The solution involves using vector representations of known intents to improve classification accuracy. The system processes user inputs by comparing them to a plurality of second vectors, each representing a known intent. Each second vector is associated with a distinct intent, such as "book a flight" or "check account balance." By analyzing the relationships between the input and these pre-defined intent vectors, the system determines the most likely intent of the user's query. The use of multiple distinct intent vectors allows the system to handle a wide range of possible user inputs and improve classification precision. The invention enhances intent recognition by leveraging vector-based comparisons, ensuring that different intents are accurately differentiated even when the input phrases are semantically close. This approach is particularly useful in applications like virtual assistants, chatbots, and customer service automation, where precise intent detection is critical for providing relevant responses. The system's ability to map inputs to specific intent vectors ensures that user queries are routed to the appropriate actions or responses.
Unknown
May 26, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.